HARE - A-MEM - On-Demand Schema Linking: leaderboard

Nested Table7, same1600 HARE reuse requests, six model columns. Fixed baseline method with on-demand schema linking (OSL), separate from its unconditional schema linking (USL) main run. USL rows repeat main scores and are not duplicated. EX is execution correctness; token count and latency excluded. Metric: Execution accuracy (%). Source: https://arxiv.org/html/2611.18338v1

Metric: Execution accuracy (%). Source: arxiv.org. Saturation forecast: Around 2029. 6 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1Gemini 4 Argon58.6#14
2GLM-5.3 Flash53.2#136
3Qwen 3.8 27B50.1#219
4Granite 4.2 8B49#1561
5GPT-6 Luna48.7#105
6Qwen 3.5 9B40.1#822

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=hare-a-mem-on-demand-schema-linking · How It Works · Data refreshed daily, snapshot 2026-11-17.